DuetGen: Music Driven Two-Person Dance Generation via Hierarchical Masked Modeling

Anindita Ghosh^1,2,3*, Bing Zhou⁴, Rishabh Dabral^2,3,Jian Wang⁴,
Vladislav Golyanik^2,3, Christian Theobalt^2,3, Philipp Slusallek^1,3, Chuan Guo⁴

¹German Research Center for Artificial Intelligence (DFKI),
²Max Planck Institute for Informatics,
³Saarland Informatics Campus,
⁴Snap Inc.

*Work done during internship at Snap Research NYC

✨ accepted at SIGGRAPH 2025 conference track ✨

DuetGen generates synchronized two-person dance choreography from input music, featuring natural and close interactions between dancers.